Can't speak for OP but I tried providing ast-grep in the execution context of an execute_bash tool, but even with pretty aggressive steering most models just don't seem to use it a lot. More expensive/SOTA models or higher reasoning increases the chances but lowers speed and raises cost. Maybe due to training bias for exploration tasks?

Yes, I've tried this passive approach too and didn't dig much further after that. I thought maybe they'd figured out something more intentional in the prompting to enable these kinds of approaches.

I have a hunch model proficiency for a given CLI tool very much correlates with how many StackOverflow answers and blog entries providing examples for it there are...

My sense is that we're at a tipping point where instruction following is getting good enough to disrupt these old habits

Not really, but interested in trying them out for a future version, especially gritql.